The problem with it is that the measurement is based on downloading a single file, with or without text, from various individual servers. That's fine for testing your speed downloading a single file, but does not accurately reflect typical behavior for web browsing. You hit a website like DU, and you're not downloading one file. You're downloading the HTML for the page itself, along with all the graphics linked to it as well as any links (web ads) to which it points, etc. This can conceivably mean dozens of independent hits for a single hit on a site and can slow things down significantly.
One site I use to demonstrate variances in speed and to show typical behavior is this:
http://www.numion.com/There are several different kinds of tests there. The YourSpeed test is excellent for measuring your effective browsing speed to a specific country or the world. It grabs numerous small graphics from several sites repeatedly for however long you tell it to do so. The longer you let it go, the better the measurement is. This mimics the behavior of hitting a typical website that pushes sometimes dozens of links at you all at the same time.
You can also test speed to specific sites, which I find very helpful when I think my connection is messed up. Turns out it is usually just certain sites that are having a problem, and if I am bored enough to trace it, it all goes to a particular set of hubs.
To get a reliable average speed, you need to allow the site to set a cookie, allow javascript of course, and take several YourSpeed tests within the US at periodic intervals over several days. (Don't always do it at the same time of day.) The site will keep track for you and tell you how you measure against others also taking the test.
On a slight tangent, another test there called StopWatch is useful for comparing browsers. You enter a URL, and it times how long it takes your browser to render the page. The browser you use can actually affect your perceived performance because it may render one site slower than another browser would. I did this when trying to choose between Opera and Firefox as my browser. (There was no signifiant difference for the sites I frequent, so I went with the OpenSource.) IE fell far behind hitting almost every site but those with ActiveX or that were optimized specifically for IE, iow non-standard sites that I don't frequent anyway.
FWIW, anything above 2.0 is good for a 3MB connection. I have a 4MB cable connection, and I got anywhere from 3.2 to 3.5 using the test you mentioned. If I were downloading one, very large file, I would get around 4.5 to 5MB. This is how marketing for broadband providers determine their averages.
This is probably a longer answer than you wanted. :-)